Check out the new USENIX Web site. next up previous
Next: Consistency and Crash Recovery Up: Separating Active and Inactive Previous: An Adaptive Grouping Algorithm

Data Lifetimes

In order to choose the proper threshold for different workloads, we calculate the byte lifetime by subtracting the byte's deletion time from its creation time. This ``deletion-based'' method was used by [1] in which all deleted files are tracked. For considering the effects of overwrites, we measured byte lifetime rather than file lifetime. Figure 3 tells the byte lifetime of four real-world workloads in details(these traces will be described in section 3.2.1).

Figure 3: Byte Lifetime of Four Real-world Workloads
\includegraphics[width=3.2in, height=2.1in]{lifetime.eps}

From the picture, we can see the active data's lifetimes shows various behaviors in different workloads. More than 70% of the data in INS and Sitar traces have a lifetime less than 10 minutes. Around 35% of the data in RES and Harp traces have a lifetime less than 10 minutes. Since the lifetime of active data varies in different workloads, it is necessary to develop this adaptive grouping algorithm to separate active data and inactive data for different workloads.



Jun Wang 2001-10-31